Modern Statistics, Exploratory Data Analysis, and Design of Experiments
In this talk, we begin with a fascinating story, as told by Prof. David Hand of Imperial College, London, in a 2008 book entitled “Statistics: A Brief Insight (Sterling).” In that book, he explained how statistics has been transformed by computers in about fifty years (1939 - 1989) from a dry Victorian discipline concerned with the manual manipulation of columns of numbers, to a sophisticated new technology, which he named “modern statistics,” involving the most advanced of software tools.
To illustrate the sheer power, excitement, and ubiquity of this new technology for engineers to take advantage of, we will next introduce a multi-scale approach to the solution of three fundamental problems (FundaP) in structural engineering (see Proc. 1979 ASTM-NIST-NSF International Symposium on Fatigue Mechanisms, Kansas City, Missouri, U.S.A., ASTM STP 675, pp. 3-8.):
(FundaP-1) Will an engineering structure work as designed ?
(FundaP-2) How long will an engineering structure last ?
(FundaP-3) Why did an engineering structure prematurely fail without warning ?
Using the concept of “uncertainty interval”, and the classical theory of design of experiments (Fisher, 1935), we show how engineers interested in an advanced and more rigorous design of high-consequence structures need to engage modern statistics (Hand, 2008) at six levels of scale: (L-1) Micro. (L-2) Specimen. (L-3) Component. (L-4) Assembly such as a structural panel. (L-5) Subsystem such as a single floor of a high-rise building. (L-6) A complete System such as a full-scale structure.
To illustrate this multi-scale modeling concept, we introduce 4 sample applications using a powerful, English-language-based, public-domain statistical analysis software named DATAPLOT. The sample problems are:
(S-1) Uncertainty estimation of time-to-failure of an aircraft window.
(S-2) Uncertainty estimation of fracture toughness of ASTM A533 Grade B-1 steel plate using an 8-factor, 17-run fractional factorial orthogonal design of experiments.
(S-3) Uncertainty quantification of design allowables of composites in commercial jets based on experimentally verified failure criteria.
(S-4) Uncertainty estimation of time-to-failure of a 100-column steel grillage on fire.
Significance of this multi-scale approach to improving the reliability of high-consequence complex engineering systems/networks is discussed.